Chevron DownCommand-Line Reference
Chevron DownAPI Reference

Using Batched Loaders

A major feature of is the availability of a number of batched (or streaming) loaders.

The advantages and characteristics of streaming are descriped in more detail in the streaming concepts section, but the highlights are:

  • Ability to parse large data sources that exceed browser memory limits (maximum allocation limits for a single string or ArrayBuffer etc tends to be less tha 1GB in most browsers).
  • While parsing is done on smaller chunks and does not freeze the main thread.
  • data can be processed (and displayed) as it arrives over the network, rather than at the end of a long request, leading to a more interactive experience.
  • transforms can be applied incrementally to the incoming data, e.g. to cryptographically hash or decrypt data

Batches: Async Iterator based Streaming

The streaming architecture is built around ES2018 async iterators rather than the more traditional Streams. Async iterators are arguably easier to work with than streams, are consistent across browsers and Node.js, and enable a "callback-less" programming style supported by built-in JavaScript language features, such as for await (... of ...) and async function *.

Note: Stream input sources is still accepted by functions, however internally processing is done via async iterators and the output of a batched parsing operation is an async iterator that yields "batches" of parsed data.

import {JSONLoader} from '';
import {load} from '';

const data = await load(url, JSONLoader, {json: options});

The JSONLoader supports streaming JSON parsing, in which case it will yield "batches" of rows from the first array it encounters in the JSON. To e.g. parse a stream of GeoJSON:

import {GeoJSONLoader} from '';
import {load} from '';

const batches = await loadInBatches('geojson.json', GeoJSONLoader);

for await (const batch of batches) {
  // will contain a number of rows
  for (const feature of {
    switch (feature.geometry.type) {
      case 'Polygon':

Streaming Data Sources

While the primary input for parseInBatches is an async iterator many input types are supported:

  • AsyncIterable<ArrayBuffer> (i.e. the iterator must yield ArrayBuffer chunks).
  • Stream instances can be used as input to parseInBatches. An async iterator will automatically be created from the stream.
  • Response objects can also be used as input (the Response.body stream will be used).

In addition, note that applications can easily wrap many data types in a Response object (e.g. FormData, Blob, File, string, ArrayBuffer etc), which makes it possible to do streaming loads from almost any data source.

Applying Transforms


Example of using a transform to calculate a cryptographic hash:

  import {loadInBatches} from '';
  import {CRC32HashTransform} from '';

  let hash;

  const csvIterator = await loadInBatches(CSV_URL, CSVLoader, {
    transforms: [CRC32HashTransform],
    crypto: {
      onEnd: result => {
        hash = result.hash;

  let csv;
  for await (const batch of csvIterator) {}


Note that by using a transform, the hash is calculated incrementally as batches are loaded and parsed, and does not require having the entire data source loaded into memory. It also distributes the potentially heavy hash calculation over the batches, keeping the main thread responsive.